Table 1. Classification result on the ModelNet40 dataset.
| Methods | Input | OA (%) | Parameters | FLOPs |
|---|---|---|---|---|
| Other Learning-based Methods | ||||
| SO-Net [30] | 2048×3 | 90.9 | - | - |
| Point2Sequence [31] | 2048×3 | 92.6 | - | - |
| PointCNN [14] | 1024×3 | 91.7 | - | - |
| PointNet† [1] | 1024×3 | 89.2 | 3.47 M | 0.45 G |
| PointNet++(SSG) † [2] | 1024×3 | 92.4 | 1.48 M | 0.87 G |
| PointNet++(MSG) † [2] | 1024×3 | 92.7 | 1.75 M | 4.07 G |
| DGCNN† [17] | 1024×3 | 92.6 | 1.82 M | 2.43 G |
| DGCNN+Pnp-3D† [32] | 1024×3 | 92.5 | 1.93 M | 3.57 G |
| PointMLP† [33] | 1024×3 | 92.8 | 13.23 M | 15.73 G |
| DualMLP† [34] | 1024×3 | 93.1 | 14.32M | - |
| G-PointNet++† [35] | 1024×3 | 92.7 | - | - |
| Transformer-based Methods | ||||
| GBNet† [36] | 1024×3 | 92.7 | 8.79 M | 9.86 G |
| Point Transformer† [37] | 1024×3 | 91.1 | 9.85 M | 18.40 G |
| PCT† [5] | 1024×3 | 92.4 | 2.88 M | 2.32 G |
| 3DGTN † [38] | 1024×3,N | 93.3 | 5.12 M | 3.09 G |
| DCNet† [39] | 1024×3 | 92.4 | 2.21 M | 7.80 G |
| PointConT† [40] | 1024×3 | 92.9 | - | - |
| Ours | 1024×3 | 93.3 | 2.43 M | 5.60 G |
† represents open source code recapitulation network experiments on NVIDIA GEFORCE RTX 4060Ti GPU. N a represents normal vector.